On Wed, 30 May 2001, K Ho wrote:
> How does Web Polygraph determine the document hit
> ratio of a given cache? Does it do one of the following:
>
> a. There is a back channel between polygraph clients & servers
There is no back channel. Back channels may be awkward to support in
some environments. For example, when the proxy itself routing traffic
and refuses to route anything but HTTP on port 80.
> b. There are extra HTTP headers in the request/response, and the
> server returns some different HTTP header for repeated requests of
> the same object (oid).
Yes, this is how it is done.
> Tcpdump suggests (b). If so, is this documented somewhere? How
> exactly does it compute the hit ratio?
The exact HTTP headers used are not documented primarily because no
intermediary should interpret those headers. I can explain the basic
mechanism though.
When a Polygraph server receives a request with header A, it generates
a reply with a mutated version of A. A Polygraph robot can tell if the
reply it received has a mutated version of A or some other mutant. If it
is a mutated version of A, it is a miss. If it a mutated version of the
header from some other request, it is a hit. The mutation algorithm is
not important and changes from one Polygraph version to another.
In short, Polygraph robots know when their requests reach origin
servers.
The scheme relies on proxies forwarding extension headers. In general,
any production-quality proxy should do that.
Finally, please note that Polygraph refuses to make hit/miss
distinction for non-basic HTTP transactions such as IMS requests or
reloads. Hit ratio is reported based on basic transactions only.
HTH,
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:19 MDT